The current cursor caching in WT only caches URIs that start with "file:" or "simple tables" (table: URIs that have no indices or column groups). We may get better performance if we can cache "layered:" tables too. Putting a "layered:" cursor in cache will mean closing (caching) the constituent cursors: ingest and stable.

If we're already caching the constituent cursors, it's not clear how much of a boost this would be, but it probably avoids some config parsing, allocating the WT_CURSOR_LAYERED struct, looking up the dhandle. So probably enough to be noticeable for workloads that use lots of collections, rather than a small number of collections.